You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@spark.apache.org by Zili Chen <wa...@gmail.com> on 2019/03/27 23:42:14 UTC
Fwd: How Spark coordinates multi contender race on writing zookeeper?
(Also on stackoverflow)
Hi guys,
Recently I open a question[1] on StackOverflow about leader election
with ZooKeeper high-availability backend. It puzzles me for some days
and it would be really help if you can take a look or even give some
thoughts.
Copy the content to mailing list:
Spark uses Curator#LeaderLatch for leader election. And PersistenceEngine
for persistent. I'd like to know what if an old leader lost its leadership
but before got notified, a new leader started to work, in which case both
of the two contenders regarded themselves as the leader and wrote on
zookeeper. Isn't it a race condition if the new leader just create a znode
on zookeeper but the old one, which was no longer the leader, delete it.
Best,
tison.
[1]
https://stackoverflow.com/questions/55380498/how-spark-coordinates-multi-contender-race-on-writing-zookeeper